# Lesson 8: The Neurobiology of Addiction: Reward, Learning and Neuroadaptation

*Companion-podcast transcript • Sarah & Kiffer*  
*Office Hours episode to listen to after working through the lesson*

---

**Sarah:** Welcome back to Office Hours. I'm Sarah.

**Kiffer:** And I'm Kiffer. Lesson eight is about the neurobiology of addiction, so let me describe an experiment with rats, published in 1989. Researchers removed nearly all the dopamine from the rats' brains and then gave them a taste of sugar.

**Sarah:** Did the rats still enjoy it?

**Kiffer:** By the measure researchers use for liking in animals, yes. Their faces showed the same pleasure reactions as any other rat's, the relaxed, rhythmic tongue movements that human infants also make with a sweet taste. What the rats lost was the drive to go and get food. They stopped seeking it, and they stopped eating.

**Sarah:** Question one. If dopamine isn't the pleasure chemical, what is it? I still hear "a hit of dopamine" used to mean a hit of pleasure.

**Kiffer:** In this story dopamine has two main jobs, and pleasure itself depends mostly on other systems. The first job is teaching. Dopamine neurons fire when something turns out better than expected, stay at their baseline when it's exactly as expected, and dip when it's worse. That's the reward prediction error. The second is wanting, or incentive salience, the pull that makes a reward and the cues around it grab your attention. The pleasure, the liking, comes mostly from tiny hotspots that respond to the brain's own opioid and cannabinoid-like chemicals.

**Sarah:** Let's make the teaching job concrete, because the word "expected" is doing a lot of work there.

**Kiffer:** Here's a hypothetical, using the lesson's simple learning rule with a learning rate of 0.3. A cue has come to predict a reward worth ten units, so its learned value is ten. One day, the reward is sixteen units. Compare that with the very first time the cue appeared, when nothing was predicted and the reward was ten. Which of those two moments produces the bigger dopamine burst? Pause the audio if you'd like more time. Take a few seconds.

*(Pause)*

**Sarah:** The prediction error is the reward minus the prediction. For the bigger reward, that's sixteen minus ten, which is six. The very first time, it's ten minus zero, which is ten. So the first, smaller reward produces the bigger burst. And after the sixteen, the new value would be ten plus 0.3 times six, which is 11.8.

**Kiffer:** That's right. The neuron tracks how much better than predicted a reward was, so a pleasant thing that arrives on schedule stops teaching anything new. That's why Redish's 2004 model matters. A drug that releases dopamine directly, arriving through the blood, isn't compared with any prediction, so in the model it keeps producing a teaching signal long after a natural reward would have gone quiet.

**Sarah:** And the evidence on that model was split. Recordings in rats, reviewed by Keiflin and Janak in 2015, found that cocaine kept producing long-lasting dopamine increases after the rats had learned the task well, which fits the model.

**Kiffer:** It does, and the blocking experiment points the other way. In the study that Panlilio and colleagues published in 2007, a tone already paired with cocaine stopped the rats from learning about a light added later. That shouldn't happen if cocaine always produces an error, so the model captures something real and leaves something out.

**Sarah:** How do the two jobs fit together, though? Teaching and wanting sound like two different theories.

**Kiffer:** They are two theories, and researchers still debate how they fit. One way to hold both is this. Prediction errors help the brain learn which cues lead to a reward. Incentive salience is the pull those cues then exert, and in some people repeated drug use makes that pull much stronger, which is sensitization. The first explains how a particular street corner comes to mean something. The second explains why, years later, it can still pull at a person.

**Sarah:** Question two is about what happens after withdrawal. Once withdrawal is over, is the addiction over?

**Kiffer:** It's an easy assumption, because withdrawal is the most visible part. Withdrawal belongs to physical dependence, though, and that's a separate thing from addiction. A patient who takes a prescribed opioid exactly as directed after surgery can have withdrawal when it stops, and in 2013 the DSM-5 work group said that tolerance and withdrawal from supervised medical use, on their own, don't indicate a substance use disorder.

**Sarah:** And it runs the other way too. In a 2010 study of fifty-six inpatients who used methamphetamine, depressive and psychotic symptoms largely resolved within a week, while craving lasted at least five weeks. And in the 2011 smoking study, craving set off by cues was greater after thirty-five days without smoking than after seven, on some measures.

**Kiffer:** That's the incubation of craving, so the end of withdrawal starts a different kind of risk. I'd like to test the drug model from question one on this, because it makes a prediction we can check.

**Sarah:** Go ahead. I'll try to keep up with the arithmetic.

**Kiffer:** Take the lesson's second worked example, where the drug-paired cue reached a value of about 41.8 after thirty trials. Now, hypothetically, the person stops using and stays away from that cue for two months. What does the model say the cue's value is at the end of those two months?

**Sarah:** Okay. Two months is a long time, so the value should have fallen a good way. Maybe back down toward the ten that a natural reward settles at?

**Kiffer:** Look at the update rule before you answer. What makes the value change?

**Sarah:** The prediction error. The value moves by the learning rate times the error, and an error only happens on a trial, when the cue appears and something follows it. If the person never meets the cue, there are no trials. So the value is still 41.8. I let the calendar do the work, and in this model time alone does nothing.

**Kiffer:** Right, the model has no clock. Now suppose the person does pass that cue, again and again, and no drug follows. What happens to the value over the first few encounters? Take a few seconds.

*(Pause)*

**Sarah:** With no drug, there's no boost and no reward, so the error is zero minus the value. The first time, that's minus 41.8, and the value drops by 0.3 of that, to about 29.3. Then about 20.5, then about 14.3, and after the fourth encounter it's about ten. After the fifth it's about seven.

**Kiffer:** So in the model, four or five encounters without the drug bring the cue back down to where a natural reward sits. That fading is called extinction: a cue met again and again without its reward stops drawing a response. Here's the trap, though. In the model, the value can only shrink during abstinence, and only when the cue is met. The smoking study found cue-induced craving greater after five weeks of abstinence than after one, and a model with no clock has no way to produce that.

**Sarah:** So what's missing from the model?

**Kiffer:** At least two things. In a 2002 review, the psychologist Mark Bouton argued that extinction leaves the original learning in place and adds new learning stored alongside it. Which one comes out depends on context, including the place, the person's internal state and the passage of time, so a cue that has lost its pull in one setting can regain it in another.

**Sarah:** And the second missing piece is sensitization. The model treats every moment alike, while Berridge and Robinson describe a wanting system that surges with stress or excitement, so the same cue can hit much harder on a bad day.

**Kiffer:** That's right. Berridge and Robinson suggest this is also why happy life events can make a return to use more likely, since excitement amplifies the urge as well as stress.

**Sarah:** A short note before the next part, which mentions overdose: support lines, including nine eight eight and the National Overdose Response Service, are listed on the lesson page. Over the same weeks that craving can persist or grow, tolerance falls.

**Kiffer:** Yes. As Lesson five described for opioids, after a period of abstinence a dose that a person once took regularly can cause an overdose, and the unregulated supply is unpredictable. Put that next to incubation and you can see why the months after withdrawal carry real risk. A return to use in that period fits what this research predicts, and it says little about a person's character or commitment.

**Sarah:** Question three is about gambling. Between April 2025 and March 2026, people in Ontario's regulated online market placed about one hundred three billion dollars in cash wagers, the total of real-money bets, and operators kept about 4.3 billion. That share of wagers kept by operators, called the hold, is a little over four percent. If players get back about ninety-six cents of every dollar they bet, how do they lose more than four billion dollars?

**Kiffer:** Because the wagers figure counts the same money many times. A dollar that's won back gets bet again, and the house keeps its share every time it goes around. Here's a hypothetical. Someone deposits fifty dollars and plays games with a house edge of five percent, betting their winnings again until the fifty dollars is gone. On average, what's the total amount they'll have bet? Take a few seconds.

*(Pause)*

**Sarah:** Every dollar bet loses five cents on average. To lose all fifty dollars, they have to bet enough that five percent of it comes to fifty. Fifty divided by 0.05 is one thousand. So about a thousand dollars in total bets, from a fifty-dollar deposit.

**Kiffer:** That's the lesson's expected loss equation run backwards. The loss equals the total bet times the house edge, so the total bet is the loss divided by the edge. The thousand is an average. On a lucky night the money lasts longer, and on an unlucky one it's gone quickly. The loss itself stays at fifty, because the play stops when the money runs out.

**Sarah:** Let me compare that with the Lower-Risk Gambling Guidelines. Say this hypothetical person has a household income before tax of seventy thousand dollars a year. The guidelines' own table puts one percent of that at fifty-eight dollars a month. They bet a thousand dollars this month, so they're about seventeen times over the limit.

**Kiffer:** Before you settle on that, what did the people in the guideline research actually report about their spending?

**Sarah:** Right. The working group measured spending as the money people lost in a month after their winnings, their net loss, along with the share of their income they spent. So the comparison is the fifty dollars lost, which is under fifty-eight. It's about 0.86 percent of their monthly income. By comparing the total bet, I had them seventeen times over the limit when they're actually under it.

**Kiffer:** That's the trap. The guidelines' website words the limit as "Don't bet more than one percent of your household income before tax per month," which sounds like the total bet, while the 2021 report by Young and colleagues built the limit from net losses. The report even notes that people who count their wins and losses separately report higher spending than people who report the net. And the money limit is one of three. The guidelines also set no more than four days a month and no more than two types of games regularly, and the limits work only when all three are followed.

**Sarah:** And the dopamine side? Why would someone keep betting again with money that's draining away?

**Kiffer:** Go back to question one. Imagine, hypothetically, a reward of ten units that arrives on half of plays, at random. Where does the learned value settle, and what happens to the prediction errors?

**Sarah:** The value settles around five, the average payout. So each win is a surprise of about plus five, and each loss is about minus five. With a certain reward the errors die away, and with an uncertain one they keep coming, play after play.

**Kiffer:** And near misses add to that. In the 2009 study by Clark and colleagues, near misses were rated less pleasant than full misses yet raised the desire to keep playing, so wanting rose while liking fell.

**Sarah:** Which could make it sound as if everyone who plays is on the way to an addiction.

**Kiffer:** The Canadian data point the other way. In Statistics Canada's 2018 survey, 1.6 percent of people who had gambled in the past year were at moderate-to-severe risk of gambling problems. That survey came before single-event sports betting was legalized and before Ontario's online market opened, so the picture may have changed since. Kardefelt-Winther and colleagues argued in 2017 that an addiction label needs harm that impairs a person's life and persists, with other explanations ruled out first. My reading is that a reward mechanism most brains share still leads to a disorder in only a minority of the people who gamble.

**Sarah:** Question four. Imaging studies have found less grey matter, the tissue that holds most of the brain's nerve cell bodies, in the prefrontal cortex of people with addiction. Goldstein and Volkow's 2011 review reported reductions of up to twenty percent. Doesn't that show that drugs damage the brain?

**Kiffer:** Here's the question for everyone. Most of these studies compare people with and without addiction at a single point in time. Give two explanations for the difference besides the drug causing it. Take a few seconds.

*(Pause)*

**Sarah:** One is that the difference came first. Something about a person's brain before any use might make heavy use more likely. Another is that some third factor drives both, so the brain difference and the drug use share a cause. And the drug may contribute too, which is the explanation people jump to.

**Kiffer:** Those are the main three, and a single scan can't separate them. Group averages also hide a lot of variation, so an individual with addiction may have a prefrontal cortex that looks entirely typical. What you need is a study that scans people before anyone has used.

**Sarah:** There's a large one from the United States. The Adolescent Brain Cognitive Development Study recruited children aged nine to eleven at twenty-two sites between 2016 and 2018, scanned them and followed them. In a 2024 analysis of nine thousand eight hundred four of those children, Miller and colleagues found that about thirty-five percent had started using a substance before age fifteen, mostly alcohol, and even a sip counted. Most of the brain differences linked to starting, including a thinner cortex, the brain's outer layer, in prefrontal regions, appeared to be there before use began.

**Kiffer:** That's an important result, with two cautions. The differences were small on average, and starting before fifteen is a long way from addiction. It also leaves room for effects of heavy use later on. Goldstein and Volkow noted that prefrontal changes persist into long abstinence, which hints at earlier differences, and that they weren't seen in adolescents who use alcohol or cannabis, which hints at effects that build with dose. Both can be true.

**Sarah:** Which brings us to where Kiffer and I disagree, on a question that Lesson nine takes up in full. Is addiction a brain disease? Nora Volkow, George Koob and Thomas McLellan argued in a 2016 review that neuroscience supports that model. I'm more skeptical. The brain changes whenever anyone learns anything, so finding brain changes in addiction doesn't by itself make it a disease. Much of this lesson's evidence comes from rats, and the human scans are mostly snapshots. In 2015, Hall, Carter and Forlini argued that the animal and imaging evidence doesn't support the model as strongly as its advocates suggest, and that it hasn't delivered better treatments.

**Kiffer:** Those are fair points, and I'd still defend a version of the model. Cue-triggered wanting that outlasts withdrawal, stress systems recruited when the drug wears off and weaker prefrontal control are lasting changes that a person didn't choose, and they help explain how someone can sincerely want to stop and still use. In 2021, Heilig and colleagues, including researchers in Hamilton and Toronto, accepted that some criticisms have merit and replied that the brain is also where recovery happens. So I'm comfortable treating addiction as a health condition with lasting changes in the brain's systems for learning, motivation and control. I hold the word disease more loosely, because it can suggest that a person's future is fixed.

**Sarah:** I'd read the same evidence as learning that has become deeply entrenched. Lesson nine sets that learning account beside the disease, choice and biopsychosocial accounts and looks at the evidence that many people recover, often without treatment. We agree on the practical rule, though. When someone says addiction is a brain disease, or that it isn't, ask which finding they're relying on, whether it came from animals or people, and whether the design can show what came first. Then name the mechanism itself, such as cue-triggered craving that can outlast withdrawal, because that's the claim the evidence can carry.

**Kiffer:** Let's pull it together with three things to take away.

**Sarah:** First, dopamine signals that something was better than expected and gives rewards and their cues their pull. Pleasure depends mostly on other, smaller systems, so "it feels good" explains only part of why a drug keeps drawing a person back.

**Kiffer:** Second, the end of withdrawal begins a different kind of risk. Craving set off by cues can persist or grow, extinction leaves the old learning in place, and tolerance falls over the same weeks.

**Sarah:** And third, check what a number measures. Total wagers count the same dollars many times, the lower-risk money limit is best compared with what's lost after winnings, and a scan taken at one point in time can't tell you which came first.

**Kiffer:** If you'd like more practice, rework the fifty-dollar gambling problem with a house edge of ten percent and then two percent. Watch how the total amount bet changes while the loss stays at fifty dollars, then compare that loss with one percent of a hypothetical monthly household income.

**Sarah:** Next time, it's Lesson nine, Why Some People Develop Addiction and Others Do Not, where twin studies, adolescence, early adversity and Rat Park take up the question of who is most vulnerable, alongside natural recovery and the brain disease debate.

**Kiffer:** Take care, everyone.

**Sarah:** See you in Lesson nine.
